Better Interface

Review the interface as one system

A strong interface is not a stack of independent audits stapled together. Review the whole experience, let each better-* skill own its domain rules, then consolidate the evidence into one prioritized verdict.

This skill owns orchestration only. Accessibility rules belong to better-accessibility; structure to better-layout; copy to better-writing; type to better-typography; color to better-colors; visual polish and motion to better-ui. Never duplicate or override their rules here. Change-scoped review of uncommitted work, branches, and pull requests belongs to interface-review, which resolves the scope and classifies findings before handing the review back here.

Core Principles

1. Resolve Scope and Mode First

Parse the invocation as [quick|full] [scope]. The first token is a mode only when it is exactly quick or full; anything else is part of the scope. Mode defaults to full.

Infer the screen, flow, feature, or repository scope from the request and current workspace. State the resolved scope in the output.

If the requested scope is too large to inspect credibly, narrow it to one complete flow: the one the request itself centers on, or failing that the entry path every user of the scope must pass through. State the boundary and what it excluded. Never imply uninspected surfaces were reviewed.

When the request names a branch, pull request, commit range, or uncommitted changes, that is a change review, not a screen review. Say so and ask the user to run interface-review; it is user-invoked, so this skill cannot start it. Never resolve a change scope here: reading a diff, classifying findings, and expanding changed files to affected surfaces are all interface-review's, and guessing at them produces a report whose scope nobody can check.

interface-review hands the review back the other way. When it does, it supplies the resolved change scope, the affected surfaces, and a status for each finding; severity, consolidation, the cap, the output format, and the verdict stay here, under Change-Scoped Reviews below.

2. Recon Before Judgment

Identify the framework, styling system, component library, design tokens, supported viewports, and available preview or test commands. Write every fix in the project's own idiom so no finding arrives as a request to adopt a different stack. That governs the form of the fix, not whether the current code is good enough.

Then read what the project has written about its own interface: CONTRIBUTING.md, CODING_STANDARDS.md, AGENTS.md, CLAUDE.md, a design-system doc, Storybook docs, interface ADRs. Name in the output which you found, or that there are none.

Read them for leverage, not permission. A documented convention is not evidence the convention is good, and "it's in the style guide" does not retire a finding. What they change is where you report: when a guideline or shared token is the cause, report it once against that source with the components as its locations.

3. Use Domain Skills as the Sources of Truth

Before reviewing, confirm that every owning skill below is available. Load and apply every available owner. In quick mode, inspect every domain but spend depth only where the primary flow has evidence. In full mode, complete each available domain review before consolidation.

Review in this order so foundational failures are not hidden by polish:

1. better-accessibility
2. better-layout
3. better-writing
4. better-typography
5. better-colors
6. better-ui

This skill owns the final response. When a domain skill is loaded through better-interface, apply its principles and references but ignore its Reporting section and the review-output.md it points at. Use the consolidated format, shared severity, and finding cap in this file instead.

If an owning skill is unavailable, mark that domain Not reviewed, name the missing skill, and continue with the remaining domains. Do not recreate its rules from memory, substitute a neighboring skill, or claim holistic coverage.

When two skills appear to cover the same issue, assign it to the skill that owns the underlying rule and mention secondary effects in the Why cell. Report it once.

4. Require Evidence

Every finding cites path/to/file:line and shows the current implementation. If the review artifact has no source files, cite the exact screen and component. Do not report a code-level finding from visual appearance alone or a visual finding from source code alone when runtime behavior determines the result.

5. Rank by User Impact

Use one shared severity scale:

• HIGH: blocks a task, misleads the user, hides content or controls, causes data-loss risk, or creates a repeated systemic failure.
• MEDIUM: meaningfully harms comprehension, efficiency, adaptability, or consistency.
• LOW: isolated polish with limited task impact. Include only in full mode.

Within a severity, rank by reach and leverage. A token or shared-component fix outranks the same symptom in one leaf component.

Escalation triggers. Once the owning skill confirms one of these, it is HIGH on sight, not averaged down because the surface is minor and not withheld in quick mode:

• An interactive control with no accessible name.
• A keyboard-reachable control with no visible focus indicator.
• A control or path reachable by pointer but not by keyboard.
• Motion or auto-playing content that ignores prefers-reduced-motion.
• Content or a control clipped, overlapped, or unreachable at 320px width or 200% zoom.
• Body or control text whose rendered contrast pair fails its required ratio.
• State or meaning carried by color alone.
• A destructive action with no confirmation, undo, or distinct treatment.

Triggers rank above every other finding. When more fire than the cap allows, list them first and state how many findings the cap excluded; a cap may shorten a report but may never be why a blocker went unreported.

These set severity, not new rules. The owning skill still decides whether the symptom is present, and this list decides what it costs. In a change review, a confirmed Regression against a trigger is HIGH even where the same symptom would be MEDIUM as pre-existing.

6. Consolidate Systemic Findings

One root cause is one finding. List every confirmed location in the same row rather than producing a row per occurrence. Do not pad the report to reach the finding cap; a short review or no findings is a valid result.

7. Make Restraint Visible

Record candidates considered but deliberately rejected. A candidate is rejected when the owning skill permits the current implementation, evidence is insufficient, the project's convention is a defensible choice and not merely an established one, or the proposed change would add complexity without user benefit.